Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/alebgl77/tierdecay/tier-decaynpx skills add alebgl77/tierdecay --skill tier-decaygit clone --depth 1 https://github.com/alebgl77/tierdecayWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/alebgl77/tierdecay/tier-decay)<a href="https://agentmods.dev/skills/alebgl77/tierdecay/tier-decay"><img src="https://agentmods.dev/badge/skills/alebgl77/tierdecay/tier-decay.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00071 | $0.00870 |
| Opus 5 | $0.00036 | $0.00435 |
| Sonnet 5 | $0.00014 | $0.00174 |
| Haiku 4.5 | $0.00007 | $0.00087 |
Grade A, and why
tier-decay scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 73 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Tier Decay — amortize reasoning
Goal: every problem CLASS is solved at a high tier at most once. Afterwards it exists as a documented pattern executable one tier lower. Expensive tokens are capital expenditure; this protocol is the amortization schedule. The routing rubric is only the cold-start prior — the ledger is the posterior.
1. Ledger — one row per task, every tier
Append to the LOG table in .claude/routing-ledger.md:
| date | class | predicted | executed | outcome | escalations | playbook |
Class signature: 2–4 hyphenated tokens, <verb>-<object>-<surface>
(e.g. add-endpoint-rest, write-migration-postgres, fix-flaky-test-jest).
Before minting a new class, scan the LOG and reuse an existing signature —
matching power depends on signature discipline.
When a class reaches ≥3 rows, compute/update its row in the PRIORS table: its empirical default tier now OVERRIDES the scoring rubric.
2. Distill — after [T2]/[T3] success only
Distill iff the class will plausibly recur (expected ≥2 future occurrences).
Write a ≤15-line entry into the repo-playbook skill, following its
format. Distill the DECISIONS — invariants, order of operations, the trap and
its avoidance — never diffs, never secrets, never volatile business values.
A wrong pattern costs more than no pattern: when in doubt, don't distill.
3. Probe — the downgrade test
On the NEXT occurrence of a distilled class, dispatch one tier below the entry's provenance tier, with the entry quoted verbatim in the brief's CONTEXT and acceptance criteria mandatory.
- Pass →
hits +1on the entry. Athits ≥ 2, the class's default tier is permanently lowered in PRIORS and the entry'sprovenance:is rewritten to the new lower tier. Decay is iterative: the counter resets and — because a live entry outranks PRIORS (model-routing§0) — the class probes the next tier down on later occurrences (T3→T2→T1), until provenance reaches T1. - Fail → escalate normally, move the entry to QUARANTINE with a one-line failure cause. The failed tier becomes the class's floor (sticky tier); revise or delete the entry on next encounter.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 73 lines · 71 tokens per session scan A b7654f079ddf
tier-decay is a skill published in the GitHub repository alebgl77/tierdecay (1 stars, last pushed 2d ago), licensed MIT. It adds 71 tokens to every session and 870 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
aider-1h-ttl
Aider uses 5min TTL by default and works around long pauses with keepalive pings. Wire up the 1h TTL beta instead.
autonomous-lesson-learning
Autonomous loop that learns while it runs: recalls stored lessons before each iteration, saves new lessons on failures and wins, and reinforces known lessons instead of repeating mistakes. Orchestrates continuous-agent-loop patterns with the egc-memory lesson tools.
atlas-setup
Use when a repository needs Atlas setup, repair/update, AGENTS.md refresh, AI memory refresh, vocabulary cleanup, or review after major codebase changes.
continual-learning
Orchestrate continual learning by delegating transcript mining to agents-memory-updater and writing durable memory to local JSON (never root AGENTS.md Learned sections). USE WHEN continual learning, mine prior chats, maintain agent memory, or the stop hook triggers this skill.
atlas-compact
Use when AI context needs cleanup — atlas doctor context-size advisories, an oversized AGENTS.md or CLAUDE.md, bloated memory or vocabulary files, or durable docs that need compaction, relocation, or dedup.
deep-init
Generate an agent-agnostic, two-tier context layer for a codebase — a lean, always-loaded CLAUDE.md plus a deep, on-demand .ai/docs/ layer (business rules, live DB schema + ORM drift, cross-component workflows, and the WHY: ADRs + a knowledge log), every claim grounded to file:line and verified to exist. Built for…